Graph Pooling

Graph pooling plays a vital role in reducing the complexity in graph while preserving the most important structural and feature information for efficient and accurate graph-level learning. This chapter describes the fundamental concepts of graph pooling for reducing graph size. It explains what graph pooling is and why it is important for effectively learning and scaling the graph. There are two major categories of graph pooling: flat pooling and hierarchical pooling. Flat pooling methods are used for aggregating the node features into a global graph representation. Hierarchical pooling techniques, including Top-K Pooling, DiffPool, SAGPool, and MinCutPool, are explored for learning multi-level graph abstractions by coarsening the graph structure. A comparative analysis highlights the strengths and limitations of flat and hierarchical pooling techniques. The chapter concludes with a discussion on the case study in real-world applications.